Review AI usage and impact
Open Settings → AI Settings → Usage & Impact to understand how configured cloud models are being used. The same analysis is also reachable from the AI impact dashboard route.
Read the summary cards
The top cards summarize the selected period:
- Cost estimates provider spend.
- Energy estimates electricity attributable to measured cloud inference.
- CO₂e expresses the associated emissions estimate.
- Tokens counts processed model tokens.
- Requests counts inference calls.
The comparison indicator uses the previous equivalent period when one is available. A percentage increase is context, not automatically a problem: a launch-review week should not be compared blindly with an idle colony week.
Energy, CO₂e, and cost are measured for cloud models only. Local inference can consume substantial device energy without appearing in those cloud estimates, so this page is not a complete power meter for the entire system.
Change the analysis
Use the period control and arrow buttons to move through time, or jump to This month and This year. Then choose Both, By category, or By model. Within a chart, switch between per-day values, running totals, and share where available.
Category views answer “which area of work drove usage?” Model views answer “which configured model drove usage?” Use both before changing a profile: a high-cost model may be justified for one specialist category but wasteful as a general default.
Act on a surprising result
If usage jumps unexpectedly:
- narrow the period to the first affected day;
- compare category and model breakdowns;
- review automation and agent concurrency;
- inspect the inference profile assigned to the affected category; and
- verify provider-side billing before treating Lotti's estimate as an invoice.
To change the route responsible for future calls, return to Models and inference profiles.